[Feature Request]: Different ouput file formats for ML models

[NiSang05]

Description

Machine Learning model ouput in other file formats

Purpose

JASP offers an intuitive and user-friendly interface for training machine learning models, allowing users to save trained models directly to their local computers. This eliminates the need to rely on Python libraries such as scikit-learn, which, while powerful, lack the visual accessibility and ease of use provided by JASP. This feature is particularly advantageous for working with smaller, localized datasets. However, a significant limitation arises in the format used for saving models. JASP currently saves trained models exclusively in the .JaspML format. While this format works seamlessly within JASP’s prediction module, it restricts the model's usability outside the JASP environment. For example, as an educationalist, I aim to train a model in JASP, use it within Python to integrate with student data obtained from an LLM system, and make predictions based on the trained model. Unfortunately, the .JaspML format cannot be directly accessed or utilized in Python, creating a barrier to broader applicability. To enhance interoperability and expand the usability of JASP-trained models, it would be highly beneficial if JASP provided the option to save models in widely supported formats, such as Pickle (.pkl) or Joblib (.joblib). These formats are commonly used in Python and would enable seamless integration of JASP-trained models into external programs and workflows. Introducing such flexibility in the saving module would significantly increase JASP’s versatility and appeal for users who wish to combine its visual modeling capabilities with the broader functionalities of Python-based ecosystems.

Use-case

JASP's user interface is exceptionally well-designed for conducting statistical analyses, and it holds tremendous potential for machine learning applications as well. Working with JASP is a pleasure due to its accessibility and functionality. However, the software could reach even greater heights if it allowed the use of trained models outside the JASP environment. While JASP’s current features for model evaluation are already excellent, expanding its capabilities to support model deployment would significantly enhance its utility. At present, after training and testing a model within JASP, I often find myself needing to replicate the process using Scikit-learn to train the same model on identical datasets. This duplication of effort is inefficient and undermines the streamlined workflow that JASP otherwise provides. If JASP introduced functionality to export trained models in widely-used formats compatible with Python libraries (such as .pkl or .joblib), it would eliminate this redundancy and allow users to deploy models directly in external environments. Such an enhancement would position JASP as not only a powerful tool for model training and evaluation but also a practical solution for real-world deployment scenarios.

Is your feature request related to a problem?

The core issue lies in the limited data output formats available for exporting trained machine learning models in JASP. While JASP provides an excellent platform for training and evaluating models, its exclusive use of the proprietary .JaspML format creates a significant limitation. This restricts users from easily utilizing trained models outside the JASP environment, such as in Python or other machine learning workflows.

Is your feature request related to a JASP module?

Machine Learning

Describe the solution you would like

It would be highly beneficial if JASP allowed users to select multiple output formats for trained machine learning models, in addition to the existing .JaspML format. This flexibility would empower users to choose the most appropriate format for their specific use cases. The .JaspML format could remain the default for seamless evaluation of predictions within JASP, while additional formats such as Pickle (.pkl), Joblib (.joblib), or even ONNX (.onnx) could be offered for broader compatibility with external software environments. Such functionality would be particularly valuable for those aiming to integrate JASP-trained models into production workflows.

Describe alternatives that you have considered

I am not an expert in R, but I believe JASP's user interface is built using R scripts. I attempted to use ChatGPT to convert a .JaspML file into the .pmml format (including the script in the additional context). While the file conversion was performed, the resulting output was not usable. As a result, I currently have to rely on Scikit-learn to rebuild the models.

Additional context

First and foremost, thank you for developing this amazing open-source software and for the continuous innovation on this incredible project. I’ve followed its development since my time as a student at the University of Amsterdam and have been impressed by the new and exciting modules that have been added over time. It’s truly remarkable to see high-quality open-source software like this thrive in today’s landscape—kudos to the entire team for such great work!

I understand that JASP is primarily focused on statistics rather than machine learning, and I recognize that the machine learning module is still in its developmental phase. It’s exciting to imagine the potential for future enhancements, such as incorporating features like computer vision and other advanced capabilities.

While I am not an expert in all the various file types used for machine learning, perhaps someone else may already know of a quick solution to address this challenge. Below is the R conversion script provided by ChatGPT:

`
install.packages("pmml") << first install >>

library(pmml)

model_path <- "<< change path to model path >>"

model <- readRDS(model_path)

<< ONLY for the decision tree, but maybe it can loop through every option of the ML module >>
if ("rpart" %in% class(model)) {
pmml_model <- pmml(model)

<< Save the PMML model to disk >>
pmml_file_path <- "<< path to save to >>"
saveXML(pmml_model, file = pmml_file_path)

print("Model saved as PMML")
} else {
print("Model is not a decision tree or is not compatible for PMML conversion")
}`

Hope I explained everything well.

[tomtomme]

In short:
This request is about exporting .jaspML to python formats Pickle (.pkl) or .joblib.
I do not get why the workflow would not continue with JASP but with python but I am no expert here and there might be good reasons.